You already have a second brain. It just doesn’t hold its shape between sessions.

This course is about fixing that. Not by picking a smarter model — by building the environment around it. Files on disk. Separated roles. Clear stopping conditions. The discipline to restart clean when a run goes sideways.

The harness thesis sits at the center: the tools, context, and feedback loops around an AI model determine most of its useful performance. The model itself is the engine. The harness is what makes the car drive.

The course runs in three tracks:

Core — Everyone takes this. Five modules, from auditing your current setup to running agents unattended. Built on Karpathy’s agent design principles.

Advanced — For those who have the Core running. Content pipelines, multi-agent coordination, voice and video as outputs.

Applied — Domain-specific implementations: solo practice, small teams, creative work. Same architecture, different data and workflows.

Cohort 1 launches with the Execution Room cohort. It will be small. I want to get it right with people who are already thinking about this.